徐宏民臺灣大學:資訊網路與多媒體研究所邱建凱Chiu, Chien-KaiChien-KaiChiu2010-05-052018-07-052010-05-052018-07-052009U0001-1808200913442100http://ntur.lib.ntu.edu.tw//handle/246246/180739在這篇論文中，我們發現，將反轉檔案使用於以視覺文字為基礎的圖片搜尋系統會遇到的問題。在資料庫中圖片數量龐大時，反轉檔案能夠快速地取出可能是搜尋目標的部分圖片，省去在大量圖片中做搜尋所費之時間。雖然近來某些以視覺文字為基礎的圖片搜尋系統宣稱他們使用了反轉檔案技巧，但是卻一直沒有人提出明確的使用方法。們研究了數種相關於文字搜尋系統的反轉檔案技巧，並將其採用在圖片搜尋上。另外，配合圖片搜尋的特點及利用視覺文字不同的重要性，在只考慮部分視覺文字的情況下，提高搜尋速度及效果。In this thesis, we identify problems for common inverted-file indexing methods in image retrieval based on visual words. Instead of searching through a huge collection of images, inverted-file is an important indexing technique which can quickly pull out a subset of images that are likely to be correct search results. While latest works trying to use this technique, there are still no clear methodologies proposed now. e investigate and adopt text-based inverted methods for image retrieval, and also propose methods to identify subsets of visual words for largely improving retrieval efficiency and performance.Chapter 1 Introduction……………………………………………………………...1hapter 2 Related Works…………………………………………………………...6hapter 3 Inverted-file for Image Retrieval………………………………………..8.1 Boolean Inverted-file…………………………………………………………9.2 Scoring Method……………………………………………………………..12.3 Efficiency Comparison……………………………………………………...15.3.1 Turtle TAAT……………………………………………...…………….……………16.3.2 Moffat TAAT…………………………………………………………………….…..16.4 Proposed Method……………………………………………………………17hapter 4 Experiment………………………………………………………………19.1 Boolean Inverted-file………………………………………………………..20.2 Scoring Method and Efficiency Comparison……………………………….23.3 Combining Moffat TAAT and L1 Distance………………………………...25.4 Modified Scoring Method…………………………………………………..26hapter 5 Conclusion and Future Work………………………………………….28eference…………………………………………………………………………….29application/pdf883836 bytesapplication/pdfen-US反轉檔案、視覺文字圖片搜尋計分法出現頻率得分上限inverted-fileimage retrievalvisual wordscoring methodterm frequencyupper boundthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/180739/1/ntu-98-R96944031-1.pdf